def __init__(self, num_features=13):
super(Xvector_TDNN, self).__init__()
out_vector_dim = 1500
self.frame1 = TDNN(input_dim=num_features, output_dim=512, context_size=5, dilation=1)
self.frame2 = TDNN(input_dim=512, output_dim=512, context_size=3, dilation=2)
self.frame3 = TDNN(input_dim=512, output_dim=512, context_size=3, dilation=3)
self.frame4 = TDNN(input_dim=512, output_dim=512, context_size=1, dilation=1)
self.frame5 = TDNN(input_dim=512, output_dim=out_vector_dim, context_size=1, dilation=1)
self.fc1 = torch.nn.Linear(2*out_vector_dim, 500)
self.fc2 = torch.nn.Linear(500, 1)
def __init__(self):
super(decoder, self).__init__()
self.tdnn_1 = TDNN(input_dim=32,
output_dim=32,
context_size=3,
dilation=1)
self.tdnn_2 = TDNN(input_dim=32,
output_dim=32,
context_size=3,
dilation=1)
self.test = nn.Sequential(
nn.AvgPool2d((38, 1), stride=1),
nn.Linear(32, 11),
)
self.output = nn.Softmax(dim=1)
def __init__(self):
super(classifier2, self).__init__()
self.tdnn_0 = TDNN(input_dim=40,
output_dim=32,
context_size=3,
dilation=1,
stride=3)
def __init__(self, feat_dim, output_dim):
super(FrameCompressor, self).__init__()
self.frame1 = TDNN(input_dim=feat_dim,
output_dim=512,
context_size=5,
dilation=1)
self.frame2 = TDNN(input_dim=512,
output_dim=512,
context_size=3,
dilation=2)
self.frame3 = TDNN(input_dim=512,
output_dim=512,
context_size=3,
dilation=3)
self.frame4 = TDNN(input_dim=512,
output_dim=output_dim,
context_size=1,
dilation=1)
def __init__(self):
super(classifier, self).__init__()
self.tdnn = TDNN(input_dim=32,
output_dim=64,
context_size=3,
dilation=1)
self.tdnn2 = TDNN(input_dim=64,
output_dim=128,
context_size=3,
dilation=1)
self.classifier = nn.Sequential(
# nn.Linear(31*32, 31*32),
# nn.BatchNorm1d(31*32),
# nn.ReLU(inplace=True),
nn.Linear(38 * 128, 1776),
nn.BatchNorm1d(1776),
nn.ReLU(inplace=True),
)
self.output = nn.Softmax(dim=1)
def __init__(self, input_dim = 40, class_num=2):
super(X_vector, self).__init__()
self.tdnn1 = TDNN(input_dim=input_dim, output_dim=512, context_size=5, dilation=1,dropout_p=0.5)
self.tdnn2 = TDNN(input_dim=512, output_dim=512, context_size=3, dilation=2,dropout_p=0.5)
self.tdnn3 = TDNN(input_dim=512, output_dim=512, context_size=3, dilation=3,dropout_p=0.5)
self.tdnn4 = TDNN(input_dim=512, output_dim=512, context_size=4, dilation=4,dropout_p=0.5)
self.segment5 = nn.Linear(512, 1500)
self.segment6 = nn.Linear(3000, 512)
self.segment7 = nn.Linear(512, 512)
self.criterion = AdaCos(512, class_num)
self.id2spk={}
with open('speakers.txt', 'r') as f:
lines = f.readlines()
for line in lines:
spk = line.strip().split()[0]
id = int(line.strip().split()[1])
self.id2spk[id] = spk
def __init__(self,
num_inputs=1,
sincnet=True,
kwidth=641,
stride=160,
fmaps=128,
norm_type='bnorm',
pad_mode='reflect',
sr=16000,
emb_dim=256,
activation=None,
rnn_pool=False,
rnn_layers=1,
rnn_dropout=0,
rnn_type='qrnn',
name='TDNNFe'):
super().__init__(name=name)
# apply sincnet at first layer
self.sincnet = sincnet
self.emb_dim = emb_dim
ninp = num_inputs
if self.sincnet:
self.feblock = FeBlock(ninp,
fmaps,
kwidth,
stride,
1,
act=activation,
pad_mode=pad_mode,
norm_type=norm_type,
sincnet=True,
sr=sr)
ninp = fmaps
# 2 is just a random number because it is not used
# with unpooled method
self.tdnn = TDNN(ninp, 2, method='unpooled')
fmap = self.tdnn.emb_dim
# last projection
if rnn_pool:
self.rnn = build_rnn_block(fmap,
emb_dim // 2,
rnn_layers=rnn_layers,
rnn_type=rnn_type,
bidirectional=True,
dropout=rnn_dropout)
self.W = nn.Conv1d(emb_dim, emb_dim, 1)
else:
self.W = nn.Conv1d(fmap, emb_dim, 1)
self.rnn_pool = rnn_pool
def __init__(self, params, path='../../../', flag=False):
super(Embedding, self).__init__()
self.params = params
if flag == True:
word_embed = np.load(path + 'data/super/word_embeddings.npy')
else:
word_embed = np.load(path + 'data/word_embeddings.npy')
self.word_embed = nn.Embedding(params.word_vocab_size,
params.word_embed_size)
self.char_embed = nn.Embedding(params.char_vocab_size,
params.char_embed_size)
self.word_embed.weight = Parameter(t.from_numpy(word_embed).float(),
requires_grad=False)
self.char_embed.weight = Parameter(
t.Tensor(params.char_vocab_size,
params.char_embed_size).uniform_(-1, 1))
self.TDNN = TDNN(self.params)